[YARN-7108] Refreshing Default Node Label Expression of a queue does not reflect for running apps

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.7.3
• Component/s: capacity scheduler
• Assignee: Unassigned
• Reporter: Prabhu Joseph
• Created: 28/Aug/17 06:22
• Updated: 28/Aug/17 17:56

Description

Refreshing a queue's default node label expression does not reflect for the running applications.

Repro Steps:

4 node cluster, two node labels label1 and label2. label1 is Exclusive Partition with Node1 and Node2, label2 is Exclusive Partition with Node3 and Node4. A default queue whose default node label expression is label1.

1.Shutdown NodeManagers on label1 nodes Node1 and Node2
2.Submit a sample mapreduce on default queue which will stay in ACCEPTED state
3.Change default node label expression for default queue to label2 in capacity-scheduler.xml
yarn rmadmin -refreshQueues
queue's config gets reflected to label2 as shown on RM UI queue section but job still stays at ACCEPTED state
4. Submitting a new job into default queue moves into RUNNING state